What next after AZ-204
By Shank
I completed AZ-204 in May 2026. The credential is Microsoft Certified: Azure Developer Associate.
In the months after that, The exam and the renewal assessment retired on 31 July 2026.
The first step, and the feeling when it closed
For a software developer, AZ-204 was the first real entry into Azure resources for microservices and modular monoliths. AZ-900 is just the overview. You took it once you could already build applications and wanted to level up. For an SDE, this was the first proper path.
Preparing for it was a nice journey. The study guide and the certification page point at a long list of docs: Container Registry, Container Instances, App Service, Functions, Cosmos DB, Blob Storage, Entra ID, Key Vault, Cache for Redis, API Apps, API Management, Event Hubs, Event Grid, Service Bus, and Queue Storage. One resource, then the next, until a scalable full-stack application is clear. Host it, store the data, lock it down, and let the services talk.
Microsoft retired it in a wider AI solutions push. The June 2026 credentials roundup says older certifications are closing so the portfolio matches AI-era roles. The replacements are AI-shaped credentials. Azure Developer Associate goes in July 2026. The replacement is Azure AI Cloud Developer Associate, exam AI-200. Building on Azure is the core. A new name was put in its place.
AI-200 keeps most of the same syllabus
Side by side, the worry settles. AI-200 is still a developer exam. Pass mark 700. The exam is 120 minutes. The course is AI-200T00, five days. It was in beta in June 2026 and generally available in July. The certification page was last updated on 7 October 2026.
The similarities are the large part. The AI-200 study guide and the AZ-204 study guide are the same kind of work. Five areas become four. Security and monitoring sit in one block, and Functions move next to messaging.
| AZ-204, as of 14 January 2026 | Weight | Where that work sits on AI-200 |
|---|---|---|
| Develop Azure compute solutions | 25 to 30% | Containers, 20 to 25%. Functions move into the connect section |
| Develop for Azure storage | 15 to 20% | Data for AI solutions, 25 to 30% |
| Implement Azure security | 15 to 20% | Joined with monitoring, 20 to 25% together |
| Monitor, troubleshoot, and optimize | 5 to 10% | Same block. OpenTelemetry and KQL |
| Connect to and consume services | 20 to 25% | 20 to 25%. Service Bus, Event Grid, Functions |
You still host the application. Container Registry, App Service, Container Apps, and Functions are all there. Service Bus and Event Grid still connect it. Secrets stay in Key Vault, settings in App Configuration. Cosmos DB stays too: the SDK, queries, consistency, indexing, and the change feed. The job is still the back end. Ship it, wire it, secure it, and see what it is doing.
flowchart TD
Both["Same syllabus"] --> Host["Registry and App Service"]
Host --> Apps["Container Apps and Functions"]
Apps --> Bus["Service Bus and Event Grid"]
Bus --> Vault["Key Vault"]
Vault --> Db["Cosmos DB"]
Db --> New["New chapter: vectors"]
New --> Emb["Cosmos embeddings"]
Emb --> Pg["PostgreSQL pgvector"]
Pg --> Red["Managed Redis"]
The part that differs is the data store, and the vector database work on top of it.
| Data skill | AZ-204 | AI-200 |
|---|---|---|
| Cosmos DB | SDK, consistency, change feed | The same, plus embeddings and a vector similarity search |
| Blob Storage | Properties, metadata, lifecycle | Off the measured list |
| PostgreSQL | Not in the syllabus | Tables, pgvector index, retrieval with a metadata filter |
| Redis | Cache for Redis, on the doc list | Managed Redis: cache, expiry, and a vector index |
On AZ-204, storage was about 15 to 20% of the exam. On AI-200, data is the heaviest section, 25 to 30%. Cosmos DB adds embeddings and vector similarity search. PostgreSQL comes in with pgvector: tables, an index for a vector query, and retrieval with a metadata filter. That is the usual retrieval-augmented shape, as a database skill. Azure Managed Redis is the cache, with expiry, invalidation, and a vector index.
The service-by-service view is the same story. Most rows are a service you already revised. The new rows are the vector databases.
| Service | AZ-204 | AI-200 |
|---|---|---|
| Container Registry | Build and publish an image | The same, plus Registry Tasks |
| Container Instances | Run a container | Off the measured list |
| App Service | Web app, slots, autoscale, TLS | A container, plus environment variables and secrets |
| Functions | Triggers and bindings, under compute | A serverless API, under connect |
| Container Apps | Create a solution | Environment, revisions, and KEDA scale |
| AKS | Not in the syllabus | Deploy and troubleshoot from manifests |
| Cosmos DB | SDK, consistency, change feed | The same, plus vector search |
| PostgreSQL | Not in the syllabus | pgvector and a metadata filter |
| Blob Storage | Metadata and lifecycle | Off the measured list |
| Entra ID and Graph | Users, apps, shared access signatures | Off the measured list |
| Key Vault | Secrets and certificates | Rotation and retrieval |
| App Configuration | App settings, with Key Vault | Store and read configuration |
| Redis | Cache for Redis, on the doc list | Managed Redis, cache plus a vector index |
| API Management | APIs and policies | Off the measured list |
| Event Hubs | Event ingest | Off the measured list |
| Event Grid | Event-driven solutions | Filters, custom events, retries |
| Service Bus | Messages | Topics, subscriptions, dead-letter queue |
| Queue Storage | Queue messages | Off the measured list |
| Monitoring | Application Insights | OpenTelemetry and KQL |
If you already prepared for AZ-204, AI-200 is an easier preparation. Most of that table is revision. The new chapter is the vector side: embeddings in Cosmos DB, pgvector on PostgreSQL, and a vector index on Managed Redis. Same profession, one new data layer.
The other shifts are smaller than the data chapter. Containers go one step further, with Registry Tasks, KEDA, and AKS manifests. Monitoring is OpenTelemetry and KQL. A handful of services leave the measured list. Those are edges. The difference you feel while preparing is the data store.
Agents and Foundry apps are AI-103, which replaced AI-102. Running models in production is AI-300, which replaced DP-100. AI-200 is the developer associate that stands where AZ-204 stood.